Something that often comes up in AI contexts and the singularity is Universal Basic Income. When everything is automated and no one can get a job, how will people live?
Sam Altman, the head of OpenAI, recently posted a thought on Twitter/X that universal computing power may well be an alternative to traditional basic income in fiat currencies.
The clip was very short, but it got my thoughts spinning tremendously. I revised a bit of what he said (that everyone gets a piece of a specific company's AI, to instead be represented by computing power in purely general terms).
Basic premise: Computerized computing power is a clearly productive industrial commodity.
With the right hardware, you can transform electricity into a very broad range of logical functions, which is then generalized as computing power. Many industries use it, service companies use it, education and digital entertainment use it, and it is tremendously important for well-organized global communication and logistics.
But above all, AI software can also use it for an ever-widening range of applications. Each year since ChatGPT arrived has been characterized by quite fierce advances, as investment in AI development has taken off at a furious pace. Materials technology, biotechnology, education, my goodness how fast it's going if you've been paying attention to what's being released left and right. It feels like living in science fiction. Five years ago I could hardly have imagined that I would be sticking my nose in my phone gaping in amazement at new AI models being released. But it is broadly a technology that will change the world tremendously. And computing power is central to making that technology work.
Computing power is also a logistics dream. It weighs nothing (well, electrons have weight, but come on now). The hardware obviously weighs something, but the information itself to be computed, as well as the result, is in relative terms extremely easy to transport to the other side of the earth. It takes up extremely little space compared to physical commodities and their refined products, and since it is completely digital, there are no losses over large distances as with raw electricity.
Generalization: We can set aside all biases about global fiat currencies now. Because this is not a fiat currency, nor is it a cryptocurrency. Instead, it is an abstraction of computing power. The credit is written in stone, and directly correlated to a given amount of computing power according to a standardized measure, which is fully possible since information technology is built on logical and precisely quantifiable processes.
We call the new currency computation credits, to generalize the concept. The credit represents a standardized amount of computing power. I sometimes call it C (credits) in the rest of the text. The currency is directly exchangeable for generalizable computing power with certain requirements/expectations of what it should be capable of. Merchants can easily present exchange rates, data centers produce the currency by committing to provide the specified amount of computing power.
C then becomes a clearly standardized commodity, which directly corresponds to digital production capability.
Almost like precious metal coins, but tied directly to productive capacity rather than the relative value of the precious metal. The exact value against other currencies will of course fluctuate due to speculation and market trends, and the value of C may experience continuous (perhaps even high) inflation as computing power becomes cheaper.
C would also not be quite like bitcoin. In relative terms, bitcoin is a bit more of a receipt that someone has burned electricity to produce heat.
Admittedly a very legitimate receipt in terms of information. But Bitcoin itself is an extremely speculative currency, as its value is primarily based on its scarcity. Perhaps something akin to a combination of cryptonetworks and Folding@Home can be used to decentralize the currency and directly connect the market to the means of production, but there may be more efficient solutions.
Advantages: The computation credit itself is of course not useful in a purely physical sense. You can't eat it as it is. You don't get warm from it as it is (unless you're sitting on top of the computer). You don't get healthy or live longer from it as it is. But the same already applies in principle to virtually all other traditional existing currencies as well.
Sure, rice, blankets, antibiotics and exercise bikes can be counted as trading instruments to meet the above points. But it is extremely clumsy, and simply extremely inefficient if everyone is going to go around swapping bags of rice for other products.
But as long as existing infrastructure (with electricity and internet intact) exists, C still corresponds to means of production that are closely linked to broader value.
The most obvious advantages to me are the following:
-
The credit is easy to exchange for virtually anything else that money can buy, as long as the seller accepts it as payment. Since the credit corresponds to an exact productive value, the incentives are high to accept the currency, as it does not risk losing absolute productive value to the same extent as other currencies.
-
The currency gives, within certain limits, a guaranteed right to the computing power. If the credits are not exchanged for other services or products, they can be used to process one's own digital production or entertainment.
An obvious use is AI, but the credits can also be used more efficiently if used for older and more optimized processes. E.g. cloud gaming, graphic rendering, data processing and the like. But AI as a holistic concept broadens the scope of computing power. Some fees and latencies may occur, but I'm pretty convinced that serious players don't want to be known as low-quality usurers.
Those who absolutely have no idea how it works can sell, invest or save their credits. They can exchange it for beer and candy if they want. Or donate to research where credits can be used very efficiently.
Those who have any degree of technical ability can use the credits directly productively by exchanging them directly for computing power, or using them for further education where free resources fall short. -
Private individuals can invest in the means of production themselves and connect directly to a digital market in a highly automated way.
Not all processes can of course be parallelized in a way that can be efficiently processed by many weak computers. But you can specify what capacity (in terms of e.g. RAM and VRAM, or processors with specific instruction sets) is available, and if you have any additional fees (which are also paid with C). In some cases, you can take on a lot of small computations with a single graphics card. In some cases, you simply need a more hefty rig.
Traditional procurement and direct customers at data centers can of course remain in parallel with this. In those cases, you can pay with regular currencies, or with C. -
Overproduction of electricity can be used very efficiently. At certain times of the day, electricity has a negative price, and at these times, in theory, the price of computations can be lower than what the credit is standardized for.
So let's say that electricity during a given hour has a negative price. Customers with hourly metering of their electricity, who own computing hardware, can e.g. significantly lower the price of their computations in order to increase the chance of having customers during these hours. Even there, automatic price allocation can be done where the lowest acceptable interval is specified by the hardware owner, and customers who are not in a hurry with their computations (e.g. hobbyists in blender who want to render some giant scene for a low price).
Financing and incentives for implementation At the national level, the state can give tax breaks to data centers.
The tax relief then corresponds to computation credits equivalent to a very small portion of the data center's capacity (at least initially). Presumably by requiring larger data centers to give priority to computing power paid for with credits, but where only a portion of the data center's total power is allocated to such activities. If the data center wants to, they are allowed to dynamically allocate that capacity to other activities in the event that no requests are made with credits.
The data center is bound to prioritizing credit customers and may not charge any extra fee within the scope of the tied portion of its business. A credit should correspond to a very exact amount of logical computations.
The data center then gets a portion of its operations guaranteed profitable, due to the tax breaks, and the state/society gets a guaranteed access to computing power.
The entire society gets an equal access to the productive capacity that they can dispose of as they please. Some of the credits can be allocated to e.g. research institutes and government agencies, but presumably primarily to citizens to stimulate the economy and distribute resources in an equitable way.
In the event that no requests are made within the framework of the credit system, the suppliers are allowed to do what they want with the free capacity, and are thus not forced to have unproductive hardware if it becomes freed up. The data centers may charge extra fees if they provide capacity in excess of the statutory partial capacity for credit processing. But then of course they compete with other players on the market in the ordinary way.
In theory, one could allow an upper limit on that concept of allowing data centers to be completely tax-free if they committed to give 100% priority to the country's credit users, but with total freedom to dynamically allocate capacity to other purposes when demand falls short of capacity.
Application: Having a single credit would simplify some things very much. At least for the technically uninitiated. Here is a new penny, it is worth about so many dollars. A little more knowledgeable can use it more efficiently. Done. But on other levels it could be considered clumsy.
Therefore, the credit system can be refined by introducing credit classes according to which instruction sets they should correspond to. Class 1 can be the simplest. Type what an ARM or even simpler processors can handle fits there.
Class 2 includes all of C1, plus more instructions. For the sake of example, e.g. everything a modern graphics processor can handle, e.g. an RTX 4090. Class 3 has full coverage of all industrially available instruction sets.
Obviously, this is also a very simplified model. I am little aware of exactly which instruction sets exist and in what turnover they are relevant. Maybe it would be better with ten, or even a hundred.
But such credit classes allow for higher granularity within the system. In addition, it is flexible, since it is possible to render exchange rates between the credit classes based on how available the different instruction sets are. It is also theoretically possible to create separate credit classes.
A, B, C etc. can refer to memory/storage speed. The highest class would then correspond to e.g. multi-channel RAM disk, while the lowest class is at the level of e.g. magnetic tape. Or HDD maybe is a better low level. Then one or more digits can indicate available instruction sets within the class standardization. Or IEEE has even better knowledge than me on how such things are specified.
Final words: Personally, I believe that such a system would be far superior to fiat currencies and precious metals. Even if you compare it with previous solutions where the dollar was backed by gold. Yes, it requires a functioning electricity infrastructure. But if the power grid fails to such a mild degree that something like this wouldn't work anywhere in the country, then I don't have much faith that we'll fare much better with gold or bills in our pockets either. Because then it has really collapsed badly, and there is some doubt that you can get what you need even if you offer actual gold.
Furthermore, the main purpose would not be to act as a currency with universal usability. But it would give citizens a very streamlined and fairer access to one of modern society's most important industries.
The sum that each citizen receives would also be extremely flexible and directly adapted to the country's digital industrial capacity, which is generally a globally coveted resource. Yes, everyone should get an equal number of credits. Initially a very low sum. Maybe even so low that some may scratch their heads over what it's actually supposed to be good for.
Yes, it may feel silly that the rich who already have so much should get this too. But they are maybe more capable of managing the resource effectively. Yes, it may feel silly that the poor and uneducated who have neither contributed to society nor understand how to use the resource efficiently should get a share of it. But they need it the most.
Regardless of who you include, it can be considered controversial. But excluding someone based on social class will likely lead to even more administration and headaches than it might seem worth.
Therefore, it is best that each person gets exactly the same amount during each distribution. Even newborn children.
Instead of turning computing power into money, opening for corruption and bad management, and then distributing the remainder, it is better if the computing power IS the money and goes directly from the means of production to the people.
Because it is the most widely usable and most fit commodity ever to do like that with.
And now the technology exists to implement it.
No comments:
Post a Comment