Another Human-Centered AI Example

In my previous post about Human-Centered AI, I gave a very simplistic example of how a common AI paradigm could be more human centric and thus, more effective. I explained that GPS directions are very precise, but in a way that is not optimized for effective human use. The ubiquitous spell-check feature provides a better, more complex example of how a common problem could be solved more effectively from a more human-centric perspective.

We’ve all experienced some sort of autocorrect failure, sometimes humorous, sometimes disastrous, but always unexpected. When humans make spelling errors, it’s typically not because we don’t know how to spell correctly, but more often because we ‘fat finger’ the wrong letter. Spell check tools are based on the premise that we don’t know how to spell a word and assume that we actually intended to type in the wrong characters or the correct characters in the wrong order.

Autocoreect example.jpg

A more accurate, human-centered approach would optimize spell check tools to accommodate for common ‘fat-finger’ keystroke mistakes rather than just actual spelling mistakes. For instance, if we fat-finger the word ‘love’ to ‘loce,’ spell checkers should recognize the error as a mistyping error, not a misspelling error.

Spell-check tools also neglect to consider the context of the words. For instance, I often mistype ‘form’ instead of ‘from’ and common spell-check tools find no fault with this since both words are spelled correctly. Ideally, the tools should recognize which word does not fit the context and replace it with the correct one.

Another common typing mistake that I make is to mistype ‘desing’ and ‘suer’ rather than design and user. You would think that after thousands of repeated similar mistakes, the system would learn to recognize and correct these mistakes. The problem is that these tools treat each event as individual, unique events rather than learning from repeated behaviors over time. A more human-centered AI design would learn how we correct each of our repeated mistakes.

I hope this illustrates how AI tools can benefit from a more human-centric design perspective. We must incorporate Human-Centered and User-Experience design principles as an integral part of the AI design, development, and testing processes.

Over the past 40 years, I’ve witnessed a common pattern where each new technology initially evolves without due consideration of the user. After awhile, the technology itself is no longer enough to earn market share and companies finally turn to adapting the technology to the user instead of forcing the users to adapt to the technology. Let’s advocate more a more Human-Centered approach right from the start with AI.