Craig Ganssle

April 20, 2022

AI and Squirrel Food: Running A Proper POC

What does it take to run a successful proof-of-concept (POC) in artificial intelligence? 


While my company is most known for the work we've done in agriculture with our Farmwave technology, this principle will apply to any industry you want to integrate machine learning.

There's been a lot of discussion lately about what it takes to run a successful POC in artificial intelligence. But I can give two examples, both of which we worked with at Cadre.

Company A

Objective: Build a visual AI to monitor and watch for ground activity for dams in Brazil to prevent natural disaster and save lives from dam breaks and massive flooding.

Testing: There is not a lot of existing data, or photo/video examples of dams breaking to build a solid library of examples used for classifying AI models. This is a good thing since a lot of data, in this scenario, would mean too many disasters already occurred.

Approach: Build a lab environment to replicate the activity of various dam breakages and the various causes. Design a pool system to collect water and release over the dam, varying each test to mimic typical activity. In this environment, set-up cameras to capture the activity and build a library of imagery from the video, classifying visual outliers along the way.

Conclusion: Establishing a budget, and dedicating a team, and time, to a POC will result in proper data to build accurate and successful AI models to achieve the objective.

Company B

Objective: Test an existing piece of hardware to monitor harvest loss on the header, or rear of a combine on corn or soybean crops - a $2+ billion value in the United States alone.

Testing: Deploy the hardware in the environment it is meant to be used. Run several passes on corn, soybean, or both crops to determine its validity and accuracy.

Approach: Due to time constraints, lack of people dedicated to test the hardware, and no budget to allow the hardware provider to help consult, organize, and run a successful POC, we can perform the following:

  • Instead of a farm field, find a vacant parking lot
  • Since there is no access to a combine at the moment, just use someones automobile, (in this case a minivan)
  • Since we're not in a field, and we do not have any corn or soybeans, go to a store and buy squirrel feed instead.
  • Spread some squirrel feed on the parking lot, put the hardware on the minivan, and drive over it.
  • Test complete

Conclusion: The hardware does not work as advised on corn and soybean crops.

See the difference?

Company A dedicated time and resources to a POC while Company B rushed and had zero resources allocated. The irony, Company B is a multi-billion dollar corporation with thousands of customers asking for the farming technology. Company A is a limited budget government in a country where few are aware anything can be done, yet their country is working on it.

Not only do organizations owe it to themselves, their shareholders, and even their industries to integrate artificial intelligence properly, but they owe it to their customers. 

Automation, machine learning, data science are not one-size fits all. It requires careful planning and strategy to truly gain the force multiplier effect of value. Those that understand this will soon be the leaders in their industry.


Craig

About Craig Ganssle

patented inventor | author | founder & ceo at @farmwave | coffee snob | USMC veteran | Jesus follower