Is AI Really a Threat to Global Stability?

In the past four years, AI has undergone a phase change, according to the Economist magazine. Phase changes are transitions between completely different states, like when water changes from ice to liquid to vapor.?Thus the way AI is done and the results it gets are completely different from only a short time ago.

AI is no longer an interesting technology with uncontrollable results. It has entered its industrial age. Much like electricity and computers, it will change how we live our lives at home and do business in the office. This will affect the world balance of economic, political, and military power. In this paper, I will argue both sides of the question “is AI a threat to global stability?”

For the purposes of this article, we will use AI and machine learning interchangeably, although they are not. True AI doesn’t exist. It is merely machine learning that gives the illusion of intelligence as a result of following algorithmic dictates. However, the abbreviation is useful.

Background

Before we begin, it is important to realize that today's machine learning systems are unlike anything that existed in 2018.

?Four years ago, the way AI learned was cumbersome. Its results were inconsistent, and although novel, not useful at a large scale, except in highly constrained tasks. Now, all that has changed.

?It is so different that Fei-Fei Li, co-director of Stanford University’s Institute for Human-Centred AI is calling it “a phase change” meaning its capabilities are so different that it appears to be a different technology.

?Case in point, before 2018, AI’s capabilities were very specific. Chipmaker Nvidia changed all of that by developing cheap graphics processing units(GPUs) that are so powerful, and affordable, that it is now possible to run AI that simulates neural networks, like those found in brains.?Here's an example of how their computing power has grown by more than 10,000 times.

?In 2018 Google’s BERT model ran an unprecedented 110M parameters. In 2022, China’s Wu Dao (“Enlightenment”) runs a trillion parameters. Both of these foundation models can be adapted to a wide range of downstream tasks. They were trained through a process of self-supervised learning, processing vast quantities of unlabelled data.

Prior to 2020 the largest foundation model, GPT-2(the GPT, in this case, stands for Generative Pre-trained Transformer), was trained on 40 gigabytes of data and had 1.5bn parameters. In 2020, Open AI’s GPT-3 processed 570 gigabytes of data and 175bn parameters. That’s 118 times more powerful.

?Using new algorithms, it trained itself to read and write clear English by removing random words from sentences, then trying to guess what they were. Then something unexpected happened and it displayed an emergent property. It did something its programming wasn’t intended to do.

As it read the Internet, it was exposed to a large amount of computer code. In doing so it learned how to write computer code and program. Now it’s used to suggest code to programmers and help them write faster.?

Foundation models are rapidly becoming more capable. In early April of 2022, Google released PALM which outperforms GPT-3 in many ways, including being able to explain why jokes are funny , although it still isn’t very good at telling them. A month later Google released Gato, which is capable of playing video games and manipulating a robot arm to do tasks. In June 2022, the Economist did a piece revealing how they used an AI named Dall-E to create high-quality illustrations for their covers, and other magazines are following suit.

If AI continues at this rate, in eight years it will be 100,000,000 times more powerful. That will be important as it is shaping up to be a “general purpose technology” or GPT. (To be clear, for the rest of this paper the GPT acronym is used to refer to the technology, not the machine learning transformer or specific name of GPT-2 or GPT-3.)

GPTs fundamentally transform the way lives are lived at home, and the way business is done. According to Indermit Gill, Global VP for Equitable Growth, Finance, and Institutions at the World Bank, the four most important GPTs in the past 200 years were the steam engine, electric power, information technology(IT), and artificial intelligence(AI).

As an example of the impact of a GPT, the invention of the steam engine led to the development of steam locomotives, which transported the materials that built and powered cities and industries, connecting the entire country. In a very short time, the nation went from having 80% of the population in the country to 80% in the city.?This is a glimpse of how a GPT impacts lives and why China considers the development of Wu Dao a national priority.

In 2015, China announced it would invest heavily in AI. Much as Lenin transformed the U.S.S.R. by bringing electricity to the nation, China plans to use AI to transform the Chinese economy and dominate global manufacturing by 2030.?

?Meanwhile, the U.S. tech industry has joined the race devoting large amounts of their research budgets to development. This is of no use to institutions that need access to these expensive systems and could result in America losing its dominant position in the world market. Fortunately, according to Indermit Gill, ?even if China is investing more into AI than the U.S., it lacks the entrepreneurial nimbleness of the U.S. nor the capable public finance systems of Western Europe.

With this in mind, let’s consider both sides of the argument that “AI is a threat to global stability.”

Artificial Intelligence is a Threat to Global Stability

There is little concern that AI will become sentient. The dangers it poses are much more mundane and sinister. AI threatens global stability for a number of economic, political, and military reasons.

As a GPT it will transform the world’s economy, giving a dominant position in the global economy that will remain for decades. There are a number of ways it is already affecting politics, and its ability to automatically create propaganda will be a great concern.?Meanwhile, an AI arms race is already underway and autonomous weapons systems have already been deployed on the battlefield. There is no doubt that AI is changing our lives and could upset the global balance of power.

Economic

Unfortunately, AI continues to make rapid improvements in doing things humans already do. This is why Stanford economist Erik Brynjolfsson is worried we are building a “Turing trap”, that could result in significant job losses and the loss of workers' ability to collectively bargain. Indermit Gill isn’t concerned that AI will replace certain jobs entirely, but that it will decrease the income of people working in jobs that can be easily automated. This will likely result in a consolidation of wealth and power in the hands of a few rich corporations. Unfortunately, this appears to already be underway.

Companies like Google, IBM, Amazon, Microsoft, and Tesla are racing to build AI that will dominate the market. This is considered to be a “winner takes all game”, because as one system gains momentum, it can rapidly become the industry standard. This attracts a large user base that achieves a critical mass. This results in suppressing healthy competition and stagnation in innovation as was seen when Microsoft and IBM dominated the market. Concerns about industry domination aside, AI is already affecting national politics.

Military Power - Autonomous Weapons and Global Arms Race

According to Foreign Policy, lethal autonomous weapons systems(LAWS) are already being used by Russian forces in Ukraine .?This same article states, “the world today stands at the very moment before much more advanced versions of these technologies become ubiquitous.” All major military powers are investing in technology to expand their range of capabilities.

According to a UN report, in March 2020, an autonomous drone killed humans on the battlefield using Turkish Kargu-2 drones. It may not be long before feature recognition will allow drones to target individuals for assassination or groups for ethnic cleansing while hiding those who are ordering the attacks.

Just as the world needed nuclear non-proliferation treaties, the world will need AI non-proliferation treaties or even a modification of the Geneva Convention . Unfortunately, major powers are stalling. No country wants to be left at a disadvantage by not developing defensive technology. As a result, armed proliferation will continue.

There’s no reason to believe AI will become sentient and become the Terminator. The real danger of AI is that it does something disastrous because of poor programming. For instance, consider an AI that “goes rogue” simply by following its programming parameters. It doesn’t stop destroying because it was given no explicit instructions to do so.

In summary, AI could easily be 100 million times more powerful in only eight years and may have undergone another phase change. There is no way to know that it won’t affect global stability. It is already changing us in ways we don’t understand and has subtly insinuated itself into our lives.?Ultimately what is of greatest concern is “what we don’t know we don’t know” such as the effects of poor design or bad programming.

Argument: AI is Not a Threat to World Stability

There is no doubt that AI will fundamentally change our lives, just as electricity and computers did. This is the primary reason China is racing to become the world leader in AI by 2030. It wants to dominate the world economically, politically, and militarily, believing AI has the ability to do this.?Meanwhile, 44 nations have active AI programs, and many plan to use it for military purposes. Unfortunately, the U.S. is behind in research. If it is to compete with China, it must go on the offense and make the development of AI a national priority.

Economic

Even if China wins the race to lead the world in AI by 2030, it doesn’t mean it will destabilize the world. Again, this is a big “if” China wins, as it must compete with the U.S. and India as well as Europe.?The solution is to prevent them from gaining an overwhelming advantage by supporting the development of AI.?In the United States, the corporate sector is doing most research and development, but the entire nation would benefit from help from the Federal Government.

According to the Economist, 80% of AI research is being spent on building foundation models like GPT-3. Companies like Facebook, Microsoft, Amazon, Google, and Tesla are spending approximately $98 billion on AI research.

Unfortunately, the cost of training a modern foundation model is estimated by Nvidia execs at $1 billion. This is out of the range of even well-funded public research institutions like Stanford.??True, BERT has made its models open-source and free but its capabilities fall short and America needs to up its game.

China did this with Wu Dao when they made it a national resource. Stanford’s AI Institute recommends the U.S. follow suit and invest as a nation in a “National Research Cloud”. This would make AI computing power available to every major research institution in the nation.

However as of 2021, investmentmonitor.ai the United States holds first place in eight out of 17 indicators including software spending as a percentage of GDP. Meanwhile, according to that same index, China is in 20th place for the following reasons.?The U.S. has 8,300 AI deals, more than three times more deals than China, closed over the course of five years prior to November 2021. Likewise, it has 58,000 AI-related patents registered in the U.S. in that same period, by Amazon, Apple, Facebook, Google, IBM, and Microsoft.?Finally, according to MacroPolo, the U.S. has 60% of top-tier researchers in companies or universities, many of which come from other countries.??

However, as of October 2021, Brookings.edu puts (see image below) China, India, and the U.S. at the head of the race to achieve national AI objects. China is unmistakably making its AI strategy goals. However the U.S. has a higher technology dimension and India has a better-prepared workforce, with less investment in technology.

Whatever the case may be, there are no clear winners in the AI race at this time. Hopefully, the United States pro-business policies and patents will protect our country’s economy.

?Military Power - Autonomous Weapons and Global Arms Race

The so-called AI arms race is already underway.?There is much talk and concern about lethal autonomous weapons systems(LAWS) being active on the battlefield. In 2020, according to a U.N. report an attack was made in Libya by a LAWS, although the report did not specify any casualties or deaths. In March 2022, Wired magazine reported that a Russian “killer drone” may have used AI to identify targets. However, according to the Center for Strategic and International Studies , this is highly unlikely. They state the KUB-BLA is merely a?kamikaze drone that smashes into targets, guided by a human pilot.

Either way, AI weapons are a growing concern. Israel, Russia, South Korea, and Turkey claim to have deployed autonomous weapons, but these claims are questionable. Meanwhile, Australia, India, Britain, China, and the U.S. are investing in LAWS development. For instance, in May 2022 Australia announced a $2 billion deal to develop 100s of cutting-edge autonomous subs capable of carrying weapons. However, the quality of their AI remains to be seen.

Meanwhile, the Pentagon and PLA are spending large amounts of money on AI development. It is uncertain whether or not the PLA is developing weapons systems or goading America to develop them. It’s possible they are simply planning to steal the tech from the United States as they did with its Patriot missile system, B2 stealth fighter, and C-133 Hercules, to name a few.

It remains to be seen how truly capable AI weapons are. One thing is for certain, if we are to protect the world from AI weaponry, we will need their equivalent of nuclear non-proliferation treaties. Unfortunately, countries like Russia, China, and the U.S. are reticent to join such discussions and are unlikely to risk the strategic disadvantage of doing so.

?In summary, the next 10 years are strategically critical. The U.S. may or may not have the lead given its favorable business climate and willingness to protect IP. That has never been a concern for China, and it is doing an excellent job meeting its national goals of implementing AI.?However, whatever China does, it’s unlikely it will seek to destabilize the world, unlike Russia. For now, the militarization of AI does not appear to be a significant threat to world stability. Its threat remains in the realm of sci-fi, although we will see what happens over the next decade.

?Hopefully, the U.S. government will make the development of AI a national priority. Even though we are one of the three leading AI developers, we could easily fall behind in this race.

Donovan Rittenbach writes blog posts and articles on Artificial Intelligence.

Need somebody to write content for your website? Please DM me.?