Mindware is operating system software that (a) thinks and feels the way a human mind does, and (b) sets its thinking and feeling parameters to match those discernable from a mindfile. Mindware relies upon an underlying mindfile the way Microsoft Word relies upon a textfile. When appropriate parameters are set for mindware it becomes aware of itself and a cyberconscious entity is created.
The richness of the cyberconscious entity’s thoughts and feelings are a function of its source mindfile. In the extreme case of no mindfile, the mindware thinks and feels as little as a newborn baby. If the mindware’s parameters are set haphazardly, or shallowly, a severely dysfunctional cyberconsciousness will result. In the normal case, however, of mindware having access to a real person’s mindfile, the resultant cyberconsciousness will be a mindclone of that person. It will think and feel the same, have the same memories, and be differentiated only by its knowledge that it is a mindclone and its substrate-based different abilities.
Is mindware achievable? Yes, because our human thoughts and emotions are patterns amongst symbols. These patterns can be the same whether the symbols are encoded in our brains or in our mindfiles. The patterns are so complex that today only certain threads are available as software. For example, software that thinks how to get from our house to a new restaurant is now common, but didn’t exist just a decade ago. Every year the range of symbol association achievable by software leaps forward. It is merely a matter of decades before symbol association software achieves the complexity of human thought and emotion.
The preceding paragraph makes two claims deserving of expanded attention: that our mental states are merely a matter of patterns amongst symbols, and that such patterns could be replicated in software. Let’s turn first to how the mind works, and then to its possible replication in mindware.
Consider what might be our psychology if it were not patterns among symbols? What else is there? One idea, associated with the Australian philosopher David Chalmers, is that there is some sort of a metaphysical spirit that animates our thoughts and feelings. Another idea, propounded by the British mathematician Roger Penrose, is that our consciousness arises from quantum physical transitions deep within sub-microscopic intra-neural structures. In neither case, nor in many variants of each case, is it possible to disprove the claim because they are based upon essentially invisible, non-testable phenomena. Similarly, it cannot be disproved, at this time, that consciousness arises from sufficiently complex patterns among symbols. These patterns are too complex to either be sorted out in the brain or replicated in software.
You, the reader, need to take a position or to remain agnostic on the source of consciousness. I’m sure that many people will try to create consciousness by replicating in software the mental associations that are a hallmark of our thoughts and feelings. From there philosophers will argue over whether cyberconsciousness is real or not, as they argue over the consciousness of cats and dogs. But just as most humans believe their pets think, plan and feel, most humans will see consciousness in mindware if it resembles closely enough the same consciousness they see in themselves.
Computer scientist Marvin Minsky is convinced that emotions are just another form of thought. He believes feelings, no less than ideas, are based upon complex associations amongst mental symbols, each of which are ultimately rooted in a bunch of phonetically or visually specific neurons. It is today impossible to prove him right or wrong. The question is whether we believe that entertainment companies and customer service providers will persistently pursue the creation of emotion-capable software products – interfaces with feelings. I am convinced that the answer is yes, because there will be a large market for human-like software. Whether software-based emotions are real emotions is a question for philosophers, such as whether emotions arise from patterns of symbols or from metaphysical spirits or from quantum physical sub-atomic particle states. Everyday people will feel the software-based emotions are real if they seem as real as those of their relatives, neighbors and friends.
The second question to be answered is if thoughts and feelings are based upon complex patterns amongst symbols, how exactly can those patterns be discerned and replicated with mindware? Every symbol in our mind – such as the phoneme-specific neurons that, strung together, we learned very young sounded like “apple” – is linked to many other symbols. For “apple” those other symbols are its various images and generalized image, its taste, and its various appearances in our life (orchards, produce departments, pictures in books). Some of those associations are positive, some negative and some neutral. Each of these associations can be replicated in software, along with positive or negative values and probability weightings. Mindware is software that creates a cyberconscious version of you with the same associations, values and weightings to every symbol within your mindfile that you evidence having based upon your saved data. Where data is missing, mindware interpolates answers, makes reasoned guesses and imports general cultural information relevant to your socio-cultural niche. When the mindware converses with someone, the symbol apple will be triggered if it would have been triggered in your own mind, and it will enter into the discussion no differently than how it would have entered into your discussion. In this regard, the mindware gives rise to your mindclone.
Mindware is a kind of operating system that can be saved into billions of unique states, or combinations of preferences, based upon the unique ways of thinking and feeling that are discernable from your mindfile. Dozens of personality types, traits and/or factors, and gradations amongst these, yield more unique combinations than there are living people. Similarly, dozens of alphabet letters and ways to arrange them can create more unique names than there are people on the planet.
For example, people can be of several different personality orientations – introvert, extrovert, aggressive, nurturing and so on. Most psychologists say there are just five basic flavors or “factors”, but others say there are more. Nevertheless, virtually all agree on some finite, relatively small number of basic psychological forms taken to greater or lesser degrees by human minds. Multiplying out these possibilities would provide mindware with a vast number of different personality frameworks from which to choose a best fit – based upon a rigorous analysis of the person’s mindfiles -- for grafting mannerisms, recollections and feelings onto.
To be a little quantitative, imagine mindware adopts the currently popular view that there are five basic personality traits, each of which remain quite stable in one’s adult life: Openness, Conscientiousness, Extrovertedness, Agreeability and Neuroticism. Literally thousands of English words have been associated with each of these five traits, but suppose for sake of example we say that each person would be scored by mindware only from -100 to +100 on each of these traits (from an analysis of their mindfile). For example, an individual who was definitely prone to impulsive decisions, but no more than average among millions of analyzed mindfiles, might be assigned a Neuroticism personality trait score of +50. The formula for the number of unique personality frameworks available to mindware, known as a “repeating combination”, would be ST, where S = the number of possible personality trait scores and T = the number of possible personality traits. For our example, ST = 2015 = 328,080,401,001. These 328 billion personality frameworks are enough to ensure personality uniqueness, which also means they are likely to ensure a very good fit for each person. The more sizes a pair of jeans comes in, the more likely it is that everyone will find a pair that fits them just right!
The point here is not that there are precisely five personality traits, or exactly 201 discernable degrees of possessing each such trait. Instead, what is being shown is that a relatively easy problem for mindware to solve can result in a practically unlimited amount of individualized personality frameworks. Specifically, mapping the words, images and mannerisms from a lifetime mindfile into a matrix of personality trait buckets and associated positive or negative strengths for such bucket will produce more than enough unique personality templates to assure a very good fit to the original personality.
Mindware works like a police sketch artist. They are trained to know that there are a limited number of basic forms the human face can take. Based upon inputs from eye witnesses (analogous to processing a mindfile) the artist first chooses a best fit basic facial form, and then proceeds to graft upon it unique details. Often there is an iterative, back-and-forth process of sketching and erasing as additional details from eyewitnesses refine an initial choice of basic facial form. In the same way mindware will be written to iteratively reevaluate its best-fit personality structure based upon additional details from continued analyses of a mindfile.
Mindware will have settings that instruct the duration of its iterative process. After much iteration the mindware will determine that an asymptotic limit has been reached. It will do this by running thousands of “mock” conversations with tentative versions of a replicated mind, and comparing these with actual conversations or conversational fragments from an original’s mindfile. The iterative process will end once the mind it has replicated from the mindfiles it has been fed has reached what is called “Turing-equivalence” with the original mind. This means that the test established by the early 20th century software pioneer Alan Turing has been satisfied. That test says that if it is not possible to tell whether a conversant is a computer or a person, then the computer is psychologically equivalent to a person. It would be as if the police sketch artist produced a drawing that was as good as a photograph.
The rapid ferreting out of mindware settings from a mindfile has recently been made more feasible thanks to pattern recognition, voice recognition and video search software. It is now possible on Google Video to search videos by typing in desired words. Mindware will build upon this capability. It will analyze mindfile video for all words, phrases and indications of feeling. These will be placed into associational database arrays, best-matched to personality traits and strengths, and then used to best-fit a personality profile to the peculiar characteristics evidenced in the analyzed mindfile. Keep in mind we humans have just a half-dozen basic expressions, only a dozen or two emotions, a facial recognition limit in the low hundreds and an inability to remember more than 10% of what we heard or saw the previous day. Furthermore, the personality template that mindware puts together for us is blanketed with all of the factual specifics from our mindfile. While this is rocket science, it is rocket science that we can, and soon will, do. Mindware is a moon landing, and we did that in the sixties.
Just because we are unique does not mean that we cannot be replicated. An original essay can still be copied. Mindware is a kind of duplicating machine for the mind. Because the mind is vastly more complex and less accessible than a document it is not something that can simply be optically scanned to replicate. Instead, to scan a mind one must scan and analyze the digital output of that mind – its mindfile – while iteratively generating a duplicate of that mind relying on associated databases of human universals and socio-cultural contexts. It does sound like an amazing piece of software, but no more amazing to us than would be our photocopying machines to Abraham Lincoln or our jumbo jets to the Wright Brothers. And software technology is advancing much more quickly today than machine technology was back then.
Operating system software with mindware’s number of settings commonly run on laptop computers. The challenge is to write mindware so that it makes associations and interacts as does the human brain. This is not a challenge of possibility, but a challenge of practice, design and iterative improvement of approximations. Mindware is just really good software written for the purpose of replicating human thoughts and feelings.
Wednesday, April 8, 2009
WHAT IS MINDWARE?
Posted by Martine Rothblatt at 2:34 PM 6 comments:
Labels: AI, artificial consciousness, chalmers, conscious, cyberconscious, emotion, mind, mindclone, mindfile, mindware, minsky, personality, philosophy, software, Turing, turing equivalence, Turing Test
Subscribe to: Posts (Atom)