7 Steps to Building “Fail Safe” AI Solutions by Developing “User Use Cases”
by Thomas B. Cross
Tap on image for 2-minute video
A recent report of AI “gone wild” in reporting hateful speech, was not the result of AI but an employee “gone rogue” making the AI report the malicious speech. This is in addition to the increasing reports, of AI systems having “hallucinations” of their own for “unknown” reasons which is even more disconcerting.
Next, you would think there is a conspiracy, coming from every corner blasting out, “if you are not running fast in the AI race, you will find extinction very soon”. Next, AI cures all corporate problems, and there is an “AI bot for that.”
Next, “if AI fails, have a team of humans, ready to “back up the AI” when it fails”.
Next, “training AI is easy like the “easy button”. Do you see the pattern in these statements ? that it seems that all the AI companies and pundits, make you feel stupid, if you are not “on the AI bus”. Then reality like any technology AI will be, long, cold, full of real global criminals using AI, against you harder and faster than you can respond or even anticipate. More than 90% of all AI projects fail because there was no real business use case must less a user use case developed beforehand with customers just buying technology and expecting to solve their business problems.
Are you prepared financially for the personal and class action lawsuits, legal fees, court costs and other financial and reputational impact to your business should a random “AI hallucination” hit your business. By having precautions in place is some “prima facie” evidence you were prepared.
Before You Build, Buy or even Use AI – Have a Plan – Here are 7 Steps for Developing “Fail Safe” AI Solutions by Thomas B. Cross
1) Start by developing AI real “user use cases” with users who really want it, not just say they do find out what they will really want to use and won’t.
2) Realize AI is about business issues, not coding, like language is not about letters but grammar. Like writing great novels and movie scripts, means writing and rewriting it, and testing the use case with a user audience.
3) Then research and evaluate AI tools. Carefully review as AI projects more often fail by selecting tools too early, then “force fitting” the use case into the tool. Importantly, consider at least two tools, to develop, test and retest until ready for release.
4) Train and retrain both the AI and users at the same time.
5) Then drive the AI hard to see if it works and breaks.
6) Then fix and put to work again on larger user groups. and then release to the general population, but importantly expect unexpected failures and have your own crisis comms team ready 7×24., as far too many AI projects fail, because the project team did not respond immediately and “effectively” to crisis situations. such as hackers, hallucinations and staff “going rogue”. Also have an ongoing audit team conduct performance analysis, compliance, legal and security testing. working together with the new features team.
7) Finally have another team start working, on new features and a roadmap for future versions, as your “thinking on AI” has changed from developing the first one.
By Thomas Cross Author of MindMeld – Merging of Mind and Metal, reviewed as the “best business book on AI”. Recipient of A I Q Award for Innovative AI Omnicomms Chatbot. Program Advisor to the University of Colorado Executive Education Program – Strategic Artificial Intelligence Program. Please contact if you need help in planning and guiding AI applications development and developing corporate AI training courses and delivery.
If you need help, contact cross@gocross.com for guidance and professional services.
