AI might generate thousands of solutions for a given problem. That means the velocity of data generation can vastly exceed the rate it can be validated by current tools. Saving Changes...
Given the strong potential for current AI tools to "make stuff up", a trust but verify approach is critical rather than following recommendations blindly. The challenges which emerge are:
- The volume of produced results (Keith's suggestion)
- The inability to provide good quality, validated input data or prompts (i.e. Garbage in - garbage out)
- The inability to staff or retain folks who have the ability to verify the outputs
Kiron Saving Changes...
Sergio Luis ConteHelping to create solutions for everyone| Worldwide based OrganizationsBuenos Aires, Argentina
The key is: what it does mean validating data produced by AI? Mainly because AI is a "parachute" that comprises lot of different artifacts. If you are talking about validating what AI artifacts like ChatGPT produces then the fail is to use it to getting a result. If you are talking about other type of things then statistics models and tools can be used to validate results. But at the end, like @Kiron stated about, the works must be done on inputs instead of outputs. Saving Changes...
"Impartial observers from other planets would consider ours an utterly bizarre enclave if it were populated by birds, defined as flying animals, that nevertheless rarely or never actually flew. They would also be perplexed if they encountered in our seas, lakes, rivers and ponds, creatures defined as swimmers that never did any swimming. But they would be even more surprised to encounter a species defined as a thinking animal if, in fact, the creature very rarely indulged in actual thinking."
