Why AI Hallucinates: The BotVerse Begins

Why AI Hallucinates

The Bot-Verse Begins.

By Mike Duke

Why AI Hallucinates

All rights reserved.

ISBN: 978-0-9836830-1-8

Printed in the United States of America.

Dedication

For Karla.

Foreword

Is it ok if an artificially intelligent algorithm (an AI bot) creates a new AI bot without human intervention? Should the new AI bot be better than the first? Should the original AI bot bear some responsibility for its child? What are the new rules? Can an AI bot follow human rules? Do human rules even apply? Should we create a new set of rules for AI bots to follow? When is it ok for an artificial entity to decide who lives and who dies?

Join me in a quick trip into the near future where artificial entities begin to represent you and every day objects around you. We will explore what they are saying about you and more importantly – why.

Meet Saga SAGA.jpg . Saga will be introducing some AI techie language along the way. Saga is not an AI bot - yet.

SAGA.jpg Hello, I’m not a bot. As far as you know.

-- Mike Duke

Acknowledgments

God – Thank you for the amazing grace.

Karla – Check!

Kyle – Seek out the bowl of petunias – oh yes, not again...

Nick - 要开心

Kenzie - 체재하다

Mila – Uh Oh!

Mom – Thank you for the never ending support.

Dad – Where’s the gravy?

Dmitri & Andrew – Friends in God forever.

Randy W. - Fearless Future Teller Maker.

Kourtney - Thank you most excellent copy editor.

Refik Anadol - Thank you for your stunning vision.

CHAPTER 1

In the beginning...

Artificial Intelligence (Fun to Fail)

If you were driving down the road at 65 mph and a small child ran out into the roadway in front of you, what would you do? You have less than 2 seconds to make a decision that will save a life, end a life, or both. Would a machine make a better decision than you can? Should it? Could it? What if there were hundreds of machines that could make the decision for you based on a preset of rules you established? Hold on for the ride – we will explore a new world coming where things will collaborate to make decsions on our behalf. Sometimes those things will miss the mark, but they will be smarter than you or I could ever be when combined. Let’s go back to the beginning and see how we started this journey to a collaborative universe of artificial intelligence.

In the ’80s, my friends and I were experimenting with AI with Duran Duran blaring on the boom box. We called the work in this space the next generation in human-robot interactions. It was a simple architecture. A person would type something in and the machine would respond.

Human: Hello, how are you?

Computer: Hello, I am fine.

The response the computer offered was a best guess based on the words the human typed in. The logic looked like the below:

If a human types a sentence that contains:

Hello, how are you?

Hello how you doin?

How are you?

Then respond with:

Hello, I am fine.

Our understanding of the question was limited to the questions we had presupposed they would ask – like the above. The response was also finite–limited simply to the number of responses that we had stored in a database (in this case Hello, I am fine.) Things might go wrong if the human typed How ya doin? We were very limited in our ability to predict what the human might say. But we got better. We added thousands of word combinations. We also began to look for key words that might give us some insight into what the human was feeling. We might even be able to guess what the human’s intent was.

But, back then, limited was really limited…your smart phone makes our computers back then look like stone tablets. Of course our AI couldn’t begin to understand all the different word combinations and their meanings - let alone the different contexts for asking these questions. Today, this science of understanding human speech is known as Natural Language Processing (NLP). That science has grown by leaps and bounds over the past few years and today characterizes AI 2.0.

But let’s go back to the late ‘80s for a moment. We realized that we could have AI do things. AI could start a program when a human asked it to. AI could listen for sounds and do stuff based on what it heard. Now, rather than just answer questions based on a complex set of rules and whatever knowledge it could derive (primitive intents), these AI subroutines could start and operate other applications. This was fun. Imagine a machine that could start applications to test other applications. We employed this technology at Lotus Development Corporation, and our bot, called Sigmond, could read its own email–and if we asked, it would reboot machines on the other side of the world in Ireland and Japan.

SAGA.jpg Chatbot : A chat robot (bot for short) is designed to simulate a conversation with human users by communicating through text chats, voice commands, or both. They are a commonly used interface for computer programs that include AI abilities.

Unfortunately, in the mid-late ‘90s, other people twisted this new kind of technology to come up with their own evil version of AI Bots. Someone created an AI that could birth its own child AI bots. Yes, this meant that programs were writing their own programs without human involvement. What was the new name for this kind of AI?  We called them Viruses. This was a HUGE fail (unless you had stock in a virus removal