“As perfect as possible AIs” are needed, not cheap sales gimmicks

Monica Lewinsky (she/her) @MonicaLewinsky  Doctors use problematic race-based algorithms every day. Why are they so hard to change? – STAT https://apple.news/AbFJCF8d6TPOLH2lF7TGvZQ
Replying to @MonicaLewinsky

It is a good thing to aim for “attentive, caring, individual, personalized”. It is not the use of methods or writing things down to help guide one self and others. It is the static nature of paper rules that require intelligent caring experienced humans to use.

Paper technologies that were taken without modification to the Internet are tied to human memorization. Tied to a false assumption that human memory is the only place to store knowledge. Every law, rule, guideline, best practices, faqs, regulation, standard on the Internet is out of date and incomplete when posted. Because of a “dump it on the web and it is public and open” mentality.

It will always go out of date. Obsolete, out of date, untraceable, unverifiable are very large problems on the Internet, and that means in all human knowledge where 5.4 Billion people now are using the Internet. And some of those are using it for the other 2.8 Billion not directly able to access the Internet.

Doctors aiming to create solutions that are specific to the individual are not the problem. It is the way they and their organizations use and “share” the knowledge where it gets corrupted, misused, abused, manipulated, controlled, over monetized. Dumping “guidelines” on humans to interpret and apply is wrong. Humans simply do not have enough lossless memory. And the cheap AIs won’t do because they do not record what they do, are not auditable, verifiable, traceable and not part of a global open process.

There are ways to verify human memory, and even to record what human brains have remembered. They are still being improved. Even with decades of work, the critical technologies and tools are not ready yet.

But it is possible to link the entire Internet, the entire body of all human knowledge, in all human languages, in all forms — to inform and assist, guide and make responsive all human organizations and processes.

Will that happen? Most likely not. But I have spent more than 50 years of my 75 years watching and encouraging. I hope in a hundred years my descendant might enjoy “scan and fix this individual once a day”, not “wait until the body crashes then charge the most to repair”.

You (or an AI carefully train for that) can go to all places on the Internet where health is tied – then verify, losslessly index, integrate globally, continuously improve, guide and apply. It is not hard but it takes levels of care, attention, and dedication beyond unaided human abilities. That is why “as perfect as possible AIs” are needed, not cheap sales gimmicks.

That level of perfection can be specified measured and approached deliberately.

Filed as (“As perfect as possible AIs” are needed, not cheap sales gimmicks)

Richard Collins, The Internet Foundation