“Possible Minds” 25 ways of looking at AI edited by John Brockman

“Possible Minds” 25 ways of looking at AI edited by John Brockman




One of the best books on AI, right now in 2019, not because of its technical deep views but because it presents many (25) arguments about different aspects of AI and why there cannot be one unified vision or view. It is one to read, just because at the end you have more questions you don’t have answers to; which is a Richard Feynman quote in parts. It does allow you to explore your own views about AI, you will align to different parts of the viewpoints presented, indeed you will create a mashup of them all where you feel comfortable.

The take away from the book is the framework and not the content from reading this book, the challenge laid down to keep up with the 25 themes are they develop, morph, link, combine, fraction and fork; especially the ones you find are at conflict with your own view and belief will be the most difficult, and there are lots.

Below is for me, some personal, interruption and thinking that means I have more to explore, and I am sorry each one is about an area that needs to be explored further. If you want answers about AI, you have been born in the wrong age. My suggestion is to ask if you can come back in say 100 years. This is the time for pioneers not town planners. If you want to be part of forming, for two or three generation times, views for them to explore in more depth; then this is your book. Pick up and read every page.

Intelligence, form when do you come and where do you go to hide?

PICK Intelligence OR {chaos, randomness, time}

/* do we see a pattern and call is intelligence, if you do it enough times does something happen means Intelligence, given enough time everything is possible.

IF OUTCOME (Intelligence) is a_process_to:

TEST: LIST {understand itself, create order, re-create itself, create something else, create something new}

 If TEST is explainable PICK {chaos, randomness, time} THEN

OUTPUT is INTELLIGENCE 

 IF OUTPUT = INPUT

INTELLIGENCE is a_process

NEXT QUESTIONS LIST {where does curiosity come from?, why do complex systems look intelligent ?}


END

Humanity is the best of our faults!

DEFINE

#MACHINE_WEAKNESS = LIST {emotions, faults, error, mistakes, misunderstandings, bias}

#HUMAN_WEAKNESS = LIST {repeat, know, exact, precise, fact, defined}

TEST FOR each MACHINE_WEAKNESS

TEST LIST {diversity, creativity, love, understanding, innovation, change, adoption, risk, compromise}

IF TEST = TRUE; THEN

         HUMANITY_STRENGTH = MACHINE_WEAKNESS

TEST FOR each HUMAN_WEAKNESS

TEST LIST{same, logic, right, copy, truth, one-outcome, efficiency, effective}

IF TEST = TRUE; THEN

         HUMANITY_WEAKNESS = MACHINE_STRENGTH

          IF ERROR (HUMAN_WEAKNESS = COPY) FOREVER THEN

           HUMAN_WEAKNESS = HUMAN_STRENGTH

END

How many AI do you need for Dystopia and Utopia?

/* The thinking here is that your view of outcome for AI depends on how many AI you believe there will be and how many code bases for AI there will be.


DEFINE AI AND {able to choice *the* outcome, is general, writes the Turing test}

RUN scenarios AI (one, few, many)

/* scenarios for how many AI are there and how many codes bases for those AI’s

If AI (one)

TEST {is there one or more code bases for the AI}

   IF one = MACHINE WILL WIN, RUN FOR THE HILLS

        DYSTOPIA = TRUE

    IF few = MACHINE WILL WIN, RUN FOR THE HILLS

         DYSTOPIA = PROBABLY

    IF many = MACHINE WILL WIN, RUN FOR THE HILLS

          DYSTOPIA = POSSIBLE

If AI (few)

TEST {is there one or more code bases for the AI}

   IF one = MACHINE WILL WIN, RUN FOR THE HILLS

      DYSTOPIA = TRUE

   IF few = MACHINE CAN WIN, THINK ABOUT THE HILLS

       DYSTOPIA = POSSIBLE

    IF many = MACHINE MIGHT WIN, FLIGHT

        UTOPIA = UNLIKELY

If AI (many)

TEST {is there one or more code bases for the AI}

  IF one = MACHINE MAY WIN, YOU CANNOT RUN

      DYSTOPIA = TRUE

  IF FEW = MACHINE CANNOT WIN, ENJOY

      UTOPIA = POSSIBLE

  IF MANY = HUMANS IN CONTROL

      UTOPIA = PROBABLY

PRINT RESULTS.

END

Signals and responses

/* humans can sense and respond, very low level of gossip/ chat

/* human ability to mashup senses and experience to understand complex ideas


QUESTION TEST [is the object that the human is carrying heavy?]

TEST (human)

IF VISUAL LIST OR {bended knee, sweat, eye popping, stain, size, colour} = TRUE

IF SOUND TEST LIST OR {cry, grunt, heavy breathing, panting, expression, words} = TRUE

IF EXPERIENCE TEST LIST OR {done this before, seen this before, heard about it before} = TRUE

     MORE TRUE = OBJECT IS HEAVY

TEST (machine)

IF VISUAL LIST OR {is a human in my visible range } = TRUE

IF SOUND TEST LIST OR {is there any sound} = TRUE

IF EXPERIENCE TEST LIST OR {where is the data} = TRUE

    MORE TRUE = DON’T HAVE A CLUE

 REPEAT TEST WHEN QUESTION [is that kiss affectionate]

END

Machines and who picks the rules

We know that machines can learn to play games without having to know the rules. The old thinking was giving the machine the rules and it will play faster and better. The next step (deep learning) was the enable the machine to explore and discover our boundaries/ rules. The example here is space invaders

The assumptions here that we need to allow the machine access to the controls (to play the virtual space invaders) and that the impact of the choices has no harm.

1. Who decides and control that the machine can control the controls

2. What happens when there are no human controls

3. What happens when the harm is real

Comparing Apples and Oranges for Ethics

Humans have this ability to deal with risk, unknows and grey. We get boundaries which move, interruption that is fleeting and translations that vary. Agreement is never absolute.

Alas these traits don’t work for machines

"Surly ... if you know everything it becomes easier to say yes and no; when you can always find a reason why. But then the reason needs a reason?"