Will Deep Learning Be Monopolized?
Artificial neural networks are a kind of cognitive memory. They are memory, because they save information in terms of weight distributions, but they also work as processors, exchanging information between units, and updating weights - recognizing patterns and learning.
Considering that there are companies today working on new kind of processors emulating neural networks (e.g., RPUs), I would expect, that soon computers may start having something we could call cognitive memory units (CMUs), a new kind of hybrid between RAM and GPU, implementing powerful artificial neural networks.
Will GPU computing power be monopolized?
Today, GPU computing power can be easily turned into money by mining coins, solving Kaggle or stock market prediction problems without much mental effort of your own. Given their power in automating very tangible, real tasks, like coming up with strategically superior moves in global political game, there are all incentives to compete in monopolizing the deep learning by amassing the GPU computing by every superpower in the world, to execute their political agendas.
Who is going to be in control?
Today, we have hand-helds owned by individuals, but supercomputers owned by governments and large organizations, used to do all kind of tasks, including track and control individuals.
Unlike with programmable computers, someone who comes up with a better general-purpose learning system, is likely to go the way of AlphaGo — beating all other political opponents in the strategic global geopolitical game.
Once we start having CMUs, it is worth considering the consequences. People may get CMU power in their smartphones, however, considering that the vast majority of such cognitive power would belong to large organizations, it is a danger, that these organizations would be outsmarted by their own supercomputer, which then takes over the world. Unlike Bitcoin, which has alternatives (in case of >50% attack), organization that gets to own more than 50% of the cognitive power, is likely stay there un-dethroned, like one that had taken out the genie from the bottle.
Controls on GPU market
Considering that computational cognitive power is likely to be treated like commodity, the GPU power is susceptible to blind buy-ups by the establishment, and some controls are necessary to ensure equal and fair access to GPU computing power (and CMU computing power in the future), anti-monopoly laws, and safeguards to regulate computing power distribution.
In addition, something like a resolution about Universal Human Rights To Computing Resources may also be important, because with the exponentially increasing computing power, it is only a matter of time, when almost every human mental task will be possible to automate: just like today computers are better at mining bitcoin, tomorrow, they will be better at coming up with ideas and ways to control muscle movements to produce valuable goods, and thus, without human rights to computing resources, humans will be irrelevant.
Will the Internet save us?
Considering the number of people’s brains carrying energetically cheap computers — human brain, I think, a real thing that can help us retain control over the computers for a little longer, is forming a connected mind through communication power of the Internet.
Even today, we could easily create an app for people to cooperate (read books, etc.) by enabling them to share screens of smart phones, code together, and cooperate in virtual game worlds, and especially, exchange objectives and ideas about them more efficiently.
Improving the communication between people’s minds on the Internet obviously could enable them to collaborate much more efficiently, and may help to keep A.I. safe.
So, will deep learning be monopolized?
It could be, but there are steps take to decrease the likelihood of it —
- control the GPU (and CMU) power distribution and market,
- implement equal human rights to computing power (test: does everyone have enough power to mine bitcoin?),
- improve communication between people’s minds, such as smart phone screen sharing, and virtual collaborative learning technologies, direct neural interfaces.
