In several different discussions over the last few weeks, I have compared the state of AI (which I still think would be better described as a Manufactured Intelligence or MI) to the state of transportation in the nation when the Model T was first being sold.
Imagine you were there for the first sale of a “T” in NYC and you heard the purchaser say “Finally, I can drive to California”. Perhaps you would consider the fact that the car needed smoother travel surfaces than a horse did and that gave you some doubt. If you were a time traveler you might note that there were no gas stations and the hardware stores that carried fuel were not guaranteed to be conveniently placed. Perhaps you might think about the fact that the car and the engine that powered it were fairly new and largely unproven technologies.
If you were able to consider all of those things, you would be hard pressed to come to any conclusion other than that first purchaser was optimistically delusional at best and creatively suicidal at worst. However, if you were honest with yourself, you could also conclude that if absolutely everything went as well as it could, such a trip was absolutely doable. Sure, it might take you a month or two but, in theory, it would have been possible even if the failure probability was north of 90%.
Basically, that early adopter would be trying to shortcut evolution. Our cultural evolution has given us Interstate highways and a vast loose network of refueling locations. As one who has driven from East TN to CA more than once, I can assure you there were no concerns that I would not be able to find fuel or that a flat tire or other minor mechanical failure would result in the cancellation of the trip. Though the vehicles I used for the trips I took were far and away more advanced than the Model T, I would suggest that with a moderate amount of proper planning, the probability of catastrophic failure for a Model T to make that trip today is probably less than 40%. That is with leaving the car the exact same and using the evolved transportation environment.
The use case(s) being pushed or marketed for AI for your average human has them attempting to do that evolution skipping thing. To accept the uses as marketed, we would have to ignore the fact that LLMs have been delivering fallacious answers to the communication medium being mined by LLMs for answers to our questions. We would have to ignore the fact that there exists among us no few “trolls” who purposefully post wrong or misleading information on that communication medium with the primary or secondary intent of corrupting the answers an AI might deliver.
I am not a Luddite. I firmly believe that there will come a time when the way AI is being marketed will seem reasonable and will reflect their actual and honest abilities. I firmly believe that there are use cases in specific industries where the AI is currently a useful and reasonable tool. (AlphaFold anyone?) But for anything “mission critical” LLMs in their current (and past) state are a risky tool at best. They should only be used if you are willing to check their work.
Unfortunately, cars and AI are not the only places where we(humanity) are attempting to skip evolution. One such effort is particularly vexing to me. The approximate cost so far of the JWST (James Webb Space Telescope) is $10billion. That covers development, manufacture, launch, and ongoing operations. If you happen to be as much of a tech nerd as I am, you will be aware of the massive amount of science that continues to be done on the data being provided by JWST. OTOH, the NASA Artemis program cost has been estimated at approximately $100billion through 2025. Four really nice people got to take some pretty kewl photos with their mobile phones. A rough seat of the pants analysis suggests we paid ten times more for a ten thousandth of the science. Surely I cannot be the only one that sees the problem with this.
A significant percentage of the Artemis cost has to do with the fact that it carried humans and they were nice enough people that we wanted to bring them back safely. Imagine how much of that cost could have been eliminated if we did not need any life support for the mission. Imagine that we had included a number of sensors and then left the capsule/sensors in orbit around the Moon. We could be receiving scientifically valuable data for months or years to come (and sure, yea, occasionally some really pretty pictures of Earth-rise). Even if we wanted to drop another robot (or five) on the surface, according to the size and complexity, it would be vastly less expensive than Artemis.
I am not a space Luddite either. I simply believe that an awful lot of what we get from sending astronauts to space would be better done by automated systems. Astronauts are hugely important for figuring out how going to space affects astronauts. Other than that, most of what they do, not all by any stretch, could be done less expensively by automation. With the exception of physical sample recovery, the automated systems would not need to be brought back which is itself a significant expenditure. Rather than wait for a conveniently timed and placed asteroid strike, we could crash the automated craft into the surface for science as we have done on some other missions.
Once we know more about space and science advances in other arenas, we might choose to spend the money to launch humans on extended missions. For the foreseeable future however, it is my considered opinion that automated missions - whether to the Moon or other bodies or Lagrange points - make far more fiscal AND scientific sense. Humans and human habitable constructs are not yet ready to do long term work in space and I think our insistence on trying to force the issue is actually retarding our scientific advancement.
I sent everything before this paragraph to a scientist friend to make sure I wasn’t leaving anything out and was reminded that some of the things NASA and other humans do is for marketing. The engineering aspects of my background lead me to find marketing to be a necessary evil but still an evil. Without humans in the story, the argument goes, other humans will lose interest and will not want tax dollars spent on it. I cannot argue effectively against that but I can say that marketing that increases the costs by a factor of 10 over the actual project costs is something I can and will argue against. Again, my opposition is rooted in the effort to skip the orderly evolution of a given science. When it is time, when we are ready, when the science justifies the fiscal and physical risk, I am all in. For now, I think we need a lot more robotic additives before we declare the mix of space travel ingredients to be soup.
No comments:
Post a Comment