Conversation

7/ Some reasons: A) inability to implement dramatic process change, B) lack of usable data, C) lack of access to both ML, engineering talent

8/ 1 more reason: businesses today largely have structured or text data, not images, voice. NLU is the least far along of those 3 AI domains

9/ Of course, companies can (and should) certainly set out to capture more voice/images and progress NLP research

10/ This of course doesn't apply to some industries that DO rely heavily on images (e.g. radiology). But today that's a smallish subset

11/ RPA is NOT AI. It's fragile, hard to deploy, rules-based process automation tech based on GUI-level integration. More on this later...

1/2 RPA is often an 80% precursor to AI, so not surprising AI isn't making an impact in areas that don't even have basic automation

Don't agree with this

I agree with you in a math sense, but for a "build a full thing" sense, still need machine-connected data & action API hookups,

But the middle is math vs. like, screen scraping + macro recordings. And the middle determines the UX/interfaces

1/2 That's a pretty good example: first exact-match DOM record&replay, than ML generalizations (5-10 years of KDD?), and now image based

Replying to

2/2 and while fun to hack on (I did!), that leaves the other 80% of vanilla dev of building a useful product around it