Two strategies for building AI products / businesses (by Andrew ng)

This is a Andrew Ng stay 2021 An old article written in , But its ideas are not out of date . Simple translation here , For everyone

https://read.deeplearning.ai/the-batch/how-to-build-ai-products-and-businesses-two-strategies/

Build Ai products / Two strategies for business :

Build Ai products / The business needs to build goals , And how to build and make difficult choices .

I've heard of two styles ：

* Get ready 、 Aimed at 、 Shooting ： Plan and study carefully . Only invest resources when you are very confident in a certain direction .

* Get ready 、 Shooting 、 Aimed at ： Develop immediately and try to execute . In this way, you can quickly find problems and make adjustments in time .

Suppose you want to build a customer service chat robot for the restaurant :

 * Should you spend time studying the catering market and chat Robots before starting development ? When the research is clear , Then gradually develop . This reduces the overall waste of time and the risk of resources ？

 * Or do it immediately , Act quickly and adjust the direction in time when you fail ？

Both methods have their advocates , But I think The best choice depends on the specific situation .

When the execution cost is high , Get ready 、 Aimed at 、 Shooting Often better , Careful study can reveal the value of the project .

for example , If your team can brainstorm some other use cases （ The restaurant 、 Airline company 、 Telecommunication company, etc ） And evaluate these use cases to determine the most promising direction , Then it may be worth taking extra time before deciding on action .

If you can do it at a low cost , And determine whether the direction is feasible in this process , And adjust the direction in time . Get ready 、 Shooting 、 Aimed at It tends to be better .

for example , If you can quickly build a prototype product to determine whether users want the product , Then quick action is meaningful .

After reaching an agreement on the product direction , When building machine learning models as part of products , I tend to Get ready 、 Shooting 、 Aimed at .

Building a model is an iterative process . For many applications , The cost of training models and error analysis is not high .

But when choosing a direction means making expensive investment or this direction is a one-way door （ It means an irreversible decision ） when , It's usually worth spending more time in advance to make sure it's really a good idea .

Keep learning!

Andrew

KNIME The comparison applies to “ Get ready 、 Shooting 、 Aimed at ” style . When the data is relatively sufficient , Just have a new idea , It usually takes no more than one afternoon , I can simply assess whether this idea is feasible . This was used purely in the past Python、Matlab、R I can't imagine waiting for language .

Happy KNIME!