Making AI-agents safe and really productive

Great for what? For making better decisions – together. How?

It is clear to most of us who are building the distributed data based Trust Infrastructure in EU and now in over 100 countries that the organization wallets will bring streams of verified data from all sorts of issuers’ wallets to holder’s wallets for decision making of all sorts. Internal data and external data to be used also when deciding which service provider is the best to pass the verifiable and other data on to. 

So far we have been focusing strongly on getting the verifiable data issued by the public sector wallets out to business and citizen wallets and the other way around with wallet2wallet xEU reporting and job applications. Not very much of less strongly verified data.

And not very much so far about how the wallets are ideal for deploying, tasking, empowering and controlling AI-agents. A dynamic duo to say the least. AI describes it well: https://www.linkedin.com/posts/bo-harald-4768b51_any-reason-to-doubt-chatgpt-if-not-much-activity-7304005165055971329-4Ubx?utm_source=share&utm_medium=member_desktop&rcm=ACoAAABPj1oB9_D7YNYACmHvY9HioUqpuULqZCo

As to less strongly verifiable data we will finally have to seriously re-evaluate the real value of asking our customers and our staff in traditional ways. Seeing the dilemma when almost full points from one person may mean the same opinion as much less from another individual  – making also the average useless. Quotes from to my 2015 blog: 

“There is an exponentially growing need for this as the information overflow continues to grow exponentially. So does the number of devices, apps and solutions meant to make it easier. 

All taken together means less time for any one thing - and this again drives changes in the plastic brain (as eminently pictured in Kit Yarrow's Decoding the new Consumer Mind). This new brain that marketeers, politicians and all communicators are addressing does not any more accept anything long-winding, complicated or boring. Still - far too often - businesses are responding to too much content by creating more content.

> The answer to the dilemma is as such pretty obvious: be at the right place, at the right time, with the right solution and in the right device(s) - with utter simplicity and clarity. If you can be playful and fun much is won. Gamification is not bad either.

> But you should avoid the too often Yes/No or thumb up/thumb down questions. This is often tendentious and even outright dangerous (as if there would always be a 100pct right or wrong choice) for the organisation asking the question and can be very unpleasant for the customer/member/citizen.

In ZEF this has been addressed in evalution, testing, voting and online buying services with

1) a "How-sure-are-you?” slider (do not use 1-5 or numbers in the first place!),

2) by using the slider horizontally for how-bad>how good and vertically for how important - it also highlights aspects that are less important to mend and:

3) the ZScoring method (patented) that eliminates the differences between answers (in both dimensions) from perfectionists and ordinary people, optimists and pessimists, cultural differencies between regions etc. 

An average of answers -  without first eliminating different use of measures - is meaningless (as users mean different things with the same slider position or number and same things with different positions) -  and at worst a base for the wrong action. “+++danger that AI does not see this+++

So now we can help AI-agents helping us – as issuers, holders and verifiers of “hard facts” – verifiable data – both at work and at home with the same logic. And of course ask the AI agent to also pick up permissioned internal and external normalised meaning-making results of the many questionneres - to customers, staff and suppliers that we have done and will do. Pretty soon the AI-agents will suggest the next ones..