Continued from here.
It was about this point in my reading of history that I happened across a history of ‘what was the first computer?‘. That convinced me that tho the past was static, history kept changing, both new evidence and old evidence being re-interpreted by new understandings from other areas of scholarship, increasingly science and technology.
However interpreted, the general message was that predicting the future from the past was fraught with failure. Most of the organizations of the world, as most of the species in a similarly evolutionary system, were gone from the earth. Consequently, I took all examples of historical parallels with a grain of salt, they tend to support the prejudices of the author and the era.
Even without being particularly critical, the then-frequently in the news analogies and claims of parallels between African slaves in the Americas and the service robots that were being made seemed extreme to me. We knew where these robots kept their mind and what level of intelligence they had, because humans had designed it all, we could read the code that implemented their mind. Well, you could if you had a lot of time and a long life, and it was true that some of the modules were the result of Deep Learning neural networks, the product of a few thousand processor-years of contemplation of various information bases — gigabytes of matrices filled with seemingly-random numbers were a bit hard to interpret. But we designed that technology. We knew how it worked in principle and practice. Our best technology, but mere mechanism.
No matter how sophisticated the mechanism, it did not include sentience as a design goal, self-awareness was not an attribute of any element. UIs were not sentient, and so could be no more slave than a cow or horse could be.
Those animals, in the modern era at least, existed because humans had caused them to exist, their entire existence was at our expense. Despite PETA, despite so many dogs treated as beloved children, despite all the cat pictures everywhere on the net, most of humanity still made a strong distinction between animals and people, and allowed farmers to raise animals to serve people’s needs. Even dogs and other pets existed largely for people’s needs and are abandoned at will, even if the animal’s life had been that of a favored child up to the point of abandonment. Better than a favored child, most people wouldn’t let children get away with what they tolerated from pets.
Every psychologist who dealt with AIs said that animals, even ones as dumb as cows or sheep, had more claim to sentience than an AI. There is light in animal’s eyes, we can tell when they are sad, hurting, unhappy, and they can tell it about us. Animals share an evolutionary history with us, we share a biology that lets us empathize.
Nobody claimed that ‘bots could empathize with us, and no preacher was claiming we had any moral obligation to empathize with them. I saw nothing to contradict that. In fact, a well-trained sheep dog misunderstood fewer instructions than even 2nd-generation AIs with the best natural language understanding and voice analysis modules.
I was much involved with the servicebots we used in the nursery and kinder care. We had 8 to 16 parents and several bots per Tessel child, and the ‘bots were the easy interface to the various local and global databases of videos, psychological and medical test information on all the children, so I dealt with servicebots often as I visited other research groups, saw their remarkable children, and worked with them to get funding for the next experiments.
I try to be patient, but bots can try your patience. Thus a continuous stream of ‘can you believe how dumb bots are?’ jokes and stories that grew, as people are wont to evolve them in gossip’s telephone game, into stories from tall tales through new and virulent urban myths. At the time, I laughed, and nobody cringed when telling them in the presence of the ‘bots, who were continuously recording everything. I could claim I was superior because I didn’t repeat them, but in fact, I am bad at telling jokes and usually can’t remember them.
Nobody was concerned about a robot’s feelings. They were, after all, not sentient.
They were AIs! AIs had not reached sentience by any reputable Turing test, tho they had been winning more game shows of late. The best of them, the ones running our servicebots, were Watson-like text processors calling on specialized modules to provide probabilities for sub-problems such as natural language processing, for clauses being true or plausible, for the contextual assumptions and how those affect interpreting the request, statement or question, … all the various things that go into determining the meaning of inputs and how to handle a request or answer a question. More sophisticated than Watson when it won Jeopardy, by 1000s of person-years of labor and testing, but essentially Watson scaled.
These AIs were the natural language User Interface to a network of more specialized mental modules. They handled breaking up questions into pieces, with the advice of the specialized modules. They evaluated the probabilities of the clauses being true, with the advice of the specialized modules. They combined the pieces and evaluated the probability of truth, the plausibility, the flaws in the question, everything, with the advice of their specialized modules.
Yes, the AI-UIs were a specialized module in themselves, but were nothing without the underlying mental modules. Provably, the User Interfaces were in no way sentient. They were mere mediaton layers between the user and selector functions managing deeper cognitive functions, reasoning modules and search modules and taking credit for them all, when in fact they merely coordinated the various subordinant expertises. There was no depth to their understanding, it was all Chinese Room, software mechanism following instructions, branching on input, masquerading as mind.
There was no reason to feel ashamed of telling a robot joke in front of a robot. You didn’t care about telling dog jokes in front of your dog, and the dog is far more sentient!
Looking back, I marvel at how good people are at polluting their futures. Will we ever learn that we never get the questions right? Not any of the questions.
So we tested the latest software, fed the bugs back through the support network, using the bots themselves as the interface to that, of course. The ‘bots got better, gradually, we all noticed that. It didn’t seem to affect the humor however, which got worse. It was clear that a lot of it was pretty low-class stuff, deeply prejudiced against anything robotic, driven by the poverty and conflict of the times. And clear that a lot of people liked that humor, they were repeated everywhere.
The ‘bots were very protective of the children, their first priority. They good at what they were good at, tho unreliable at being good at them if things changed. They could change a diaper in the nursery fine, but were sometimes confused in doing so if the child was out of the nursery, on the lawn or at their pediatrician’s office. They never needed more than one example, but generalization was real slow for the ‘bots, and there are 1000s of things to learn to do, and 1000s of places to learn to do them. Robots would not be McGivers for some time, and people around them had adopted the attitudes of elementary school teachers, tended to micromanage the ‘bots.
Servicebots were treated as big children, when people were being nice. Sometimes people weren’t nice. It was somehow even easier to take a bad day out on a ‘bot than your spouse, and people didn’t much object.
But, ‘bots got better. So long as we had sufficient local processor power and low-latency, high-bandwidth links to a server farm, by the time the first children were 5, we would have had a hard time running things without them. They did all of the repetitive, manual labor, and did most of it without direction. Except for things changing and new problems, they managed themselves, the functions of laundry and cleaning, making beds and feeding and changing clothes and diapers were gone from our minds. We didn’t miss them.
We were a bit surprised when ‘bots began playing with the Tessels’. We thought it was cute, wondered what the Tessels liked about them. It took us longer to notice that the Tessels liked playing with the ‘bots better than with normals, parents, other adults or other children, and yet longer to notice that the most outstanding of the Tessels played with the ‘bots more than the others. There was some affinity between them, particular Tessels and ‘bots. We hadn’t realized ‘bots had any individuality before this.
Meanwhile, dealing with ‘bots remained frustrating. Our servicebots were hard to instruct, because, we learned, language is even more variable than environments. If there are 1000s of places to learn to do 1000s of things, there are 10s of billions of ways to use the 10,000 words of a minimal vocabulary in ordinary work situations. Our researchers and parents generally had rather larger vocabularies than that, and were frustrated by how many things had to repeated in how many ways.
Natural Language Processing had a long way to go, it seemed to me. I noticed that even highly intelligent people raised their voices to be better understood, despite the fact that it never helped. All of the jokes were not low-level, and the ‘bots heard them all.
About this time, our 2nd cohort of Tessels were born. We began this cycle of research secure that no Tessel had yet turned out bad, and tho none of the first cohort were obviously savants by age 5, most were obviously quite intelligent.
We still didn’t know causes of anything, this was still all based on my original hypothesis that different timings in the mixed brain cells produced the different and superior brains, and the continuing animal experiments. Those continued to produce interesting results, with startling results in 2-parent fusions of different species of Macaques. Again, we saw an exponential effect, they produced an animal with a very large brain for its species. It was very intelligent relative to adult Maques, tho yet young.
However, by this time, the animal experiments didn’t have much to teach us on the intelligence front, about this time is when the neutraceutical, vitamin and supplement and nootropic manufacturers started supporting that work, hoping to show an effect of their product in addition to the hybrid vigor of the polyploid-hybrid, mosaic or Tessel, however you might term them. This were standard kinds of experiments by now, feeding the foster mothers the drugs while the fetuses developed, then behavioral testing, physiology and anatomy optional if something is having a behavioral effect. Some of that was done in cell cultures, which made the work faster and less expensive, but was also of unknown relevance compared to a brain developing in its normal environment.
We didn’t understand the results of some human mixes that had produced both much better Tessels and somewhat worse than expected based on any of the measures of the sets of parent’s genetic differences. However, generally those were consistent with overall genetic differences and how well mixed cells were in the muscles of the head. (Added later : using striated muscles to assess this was a mistake, this article says hair follicles are derived from neural crest cells, that would be much easier. Will be fixed in the rewrite.)
We had improved handling of egg, clonal lines, inducing Polypotent Stem Cells, etc. considerably in the previous 5 years. We now had high reliability of mixing the iPSCs uniformly as they were injected into the gastrula-stage embryo, thus this group of Tessels had that big advantage, being the most vigorous possible hybrids of their parents.
Our first set of Tessels had used parents from the local research and academic communities. Those had a lot of diversity, tho not aborigines of any continent. This set had more different intelligences to mix. People had been volunteering, offering their genetics and presenting evidence of their intelligence. 23andMe could not categorize some of the applicants, and even our sequencing of their genome would only reveal them to be ‘ancient American Indian’, ‘unknown Inuit group’, or some similar category. We even found 2 new mitochondrial subhaplogroups.
It was a puzzle whether strange genetics were attracted to this project, or the highly intelligent were strange genetics. Whichever, our applicant dataset made it clear that previous assessments of the world’s genomes hadn’t searched in enough corners, cracks and crannies of the world. There were a lot of differences yet to be discovered.
We had decided that this was a safe procedure — we hadn’t seen anything to be aware of beyond the initial selection for sex of the parent. HLAs, blood types, etc all learned to tolerate each other during formation of the immune systems, although some specialists wondered if the Tessels would be more prone to autoimmune diseases. (I had an answer for that, of course.***)
So, being safe, and knowing we had minds with unusual features arising from the mixes, even tho the oldest were only 5 years old, humanity made investments in the future. Outside of our research group, which was positive-sum, open source, this stage of research was organized with parents forming corporations which publicly traded stock. All of the corporations were set up in a tax haven, for some reason.
I think if indentured servitude had been permitted anywhere in the world, they would have tied the Tessel to the corporation that way. The way it was, until the individual Tessel was “of legal majority”, their maintenance was the responsibility of the corporation and their income stream belonged to the corporation. Likewise, if the individual Tessel child invented things, the corporation owned it and royalties went to the corporation for the life of the patent. I should have seen the trap that was the phrase “of legal age”, but at the time was just amazed to see the Tulip Mania in progress. Interesting times, so many historically rare things happened in my lifetime.
The investment boom increased the production of Tessels in other research groups. Soon the population outside of our open collaboration was larger than in it. They used all of our prior results, of course, so began at the same technology. This was the equivalent of forking a project under BSD license in the open source world, and keeping the new code private. They were betting that by using the combined results of our continuing work with their individual proprietary research, that within a few years they would be significantly ahead in technology.
I thought it an unlikely winning strategy. Our big advantage was the transparency and engaged minds that made possible. They could not be fully engaged with as many sources as our researchers, because to do so would be to ask and answer questions that told us what they were doing. They were handicapped by their need for secrecy, of course, not enabled by it. All secret projects suffer from this, and thus are generally behind public R&D technological curves.
In the open source world, code was moving from proprietary to open source, as companies share the benefits in return for reduced maintenance costs. Companies routinely paid OSS programmers to upgrade the security or features of particular pieces of code. Some BSD-licensed projects were forked and the new OSS codebase supported by for-profit companies as a way of discarding an incompetent team that controlled it. A few have been for-profit taking a specialized version of the product for a more limited use. But many fewer have been successful takeovers for profit, in competition with the OSS original and its team. That is center-periphery Red Queen evolutionary dynamics, don’t be in the center.
Theirs was a high risk, high return strategy, not foolish if you understood that. They could get lucky and find a new key advantage and keep it so long as the patent or their secrecy held, using the proceeds to fund the research to stay ahead of the rest of us. Tech companies have been very successful with that strategy, provided they were the 1 in 10 that got that initial advantage, which they rarely did in competition with open source. But lightning does strike, and if you need that, best to lean against the flag pole.
My notes from the time record that this seemed to me to be a classic tech boom and bust cycle, but the major losers were going to be Tessels, who would remain much smarter than the executives and investors who used them in a negative-sum game. It promised to be an interesting dynamic. Has there ever been a legal savant?
From the much wider palette of genetics available for this cohort, our research program composed Tessels of yet greater variations in the set of parents. Our group’s cohorts were to be about 100 / year for the next 5 years, 5 times the 100 in the first 5 years.
In this set of children, I noticed the parents were selected for very high intelligence as well as genetic diversity. That intelligence was important in producing Tessels seemed to me to be an assumption, one we hadn’t tested. All of our previous parents had been a couple standard deviations above average, universities and research groups are, but none were certified genius, much less savants. Lab animals are pretty uniform, the range of intelligence isn’t large, so combining different intelligences isn’t easy to do with animals and we couldn’t do everything simultaneously with the few human subjects we had. I made a mental note that intelligence and genetic diversity were independent dimensions that needed checked independently.
Human strain hunters didn’t begin their searches until we had the earliest training results from these Tessels, still 10 years in the future. They found that searching for human landraces was a harder job, plants could be shy, but didn’t actively hide.
In psyops, the message is the op.
*Generalissimo Grand Strategy, Intelligence Analysis and Psyops, First Volunteer Panzer Psyops Corp. Cleverly Gently Martial In Spirit
***Autoimmune disease. At base, autoimmune diseases like Multiple Sclerosis, Lupus, rheumatic heart disease, Crohn’s Disease, … are due to the immune system not having real intestinal or blood parasites to deal with and finding some protein in their own body that reminds them of something like a bacteria, heart valve tissue, in the case of rheumatic heart disease. Idle leucocytes look for trouble.
So some genius researchers gave people with Crohn’s disease roundworm infections. Their bowel problems disappeared and stayed gone until the infections were killed.
It has seemed obvious to me since I read that study, that if you raised roundworms, etc (such a big set of alternatives) in vitro and extracted the proteins, which are the antigens, you could take them in a pill and convince your immune system you had an infestation, keep it from idleness. I don’t think that the proteins have to be kept in physiological conditions, surely anything that makes it through the stomach is being denatured? So non-physiological extraction and purification conditions would be OK.