AI Is NOT A Threat
Dispelling the myths and scaremongering around the "dangers" of AI research
The exponential rise in the faculties of AI in recent years has exacerbated the incessant scaremongering around the possibility of AI going rogue and spelling doom for humanity. Concerns about job safety among the masses are understandable, but the scaremongering surrounding this new technology, reflected in high-stakes lawsuits and recent blockbuster movies like Atlas, is disconcerting. Even more alarming is hearing these fears echoed by many leading scientists and tech CEOs including Elon Musk, a real-life "brillianaire" hero on a mission to save humanity. While many concerns are centered around the deadly outcomes that could arise from the use of AI by humans for purposes not so beneficial to humanity as a whole, which may be a genuine area of concern albeit no more than a knife or nuclear technology, the idea of AI itself choosing the path to harm humanity is a far-fetched and near-impossible scenario insofar as the recent strides in AI are concerned. Over the next few paragraphs, I hope to delineate the reasoning that undergirds my belief that no matter how astonishing its faculties, modern AI is not a threat to humanity. At least not on its own.
Let us first clarify our terms. The ceaseless speculation around AI’s threat to humanity falls under two main categories: threat from the use of AI against humans and the threat from an autonomous sentient AI. The fears around threats in the first category are grounded in sound evidence and supported by strong reasoning, underscoring legitimate risks and potential consequences in the form of geopolitical conflict, thereby warranting serious consideration and proactive measures. However, such risks can only be mitigated not by slowing down but by doubling down on research in AI because the threat posed by a competing geopolitical power can be counteracted only through deterrence. Assuming that this current trend can lead to superintelligence, which I personally don’t think is possible, the first to develop it will have an enormous advantage on the global scale because such a system would arm its creators with the ability to deter or dissuade all potential enemies from ever becoming a threat. Of course, the threat of the use of human-controlled albeit AI-powered attacking systems remains. However, only similar systems can effectively counteract such systems. Therefore, the anthropogenic threat from AI research should motivate us to intensify our research efforts, not to reduce them. The second category is more intriguing and will be the focal point of the remainder of this essay.
As many have pointed out, the shockingly amazing feats produced by AI that have led some to surmise that AI may have already achieved latent sentience, are simply instances of what Nobel Laureate Daniel Kahneman calls “System 1” intelligence at play. In fact, much of the AI research that has produced these feats has come from a technique called “deep learning”, a technique modeled after the learning patterns of the human brain. However, it is primarily the lower-order intelligence that we have been able to imitate in computer systems i.e. the intelligence that is wired into us, the intelligence that powers muscle memory or survival instincts in humans, the systems in place that allow us to learn and improve through memetic repetition; the “System 1” brain. The efforts to replicate the higher-order “System 2” type intelligence, which allows us to do complex tasks and use skills learned through System 1 like language, counting or painting to produce something truly meaningful and pursue more abstract ideals, have seen negligible progress with regards to artificial intelligence.
One might wonder how some of the smartest people on the planet with access to a virtually limitless supply of cash and having created such incredible AI that is better at imitating human greatness than humans themselves, have failed to produce a system that can perform the most fundamental activity we call “thinking”. Though these systems are often presented in various sci-fi narratives in popular culture, their speculated timeline is often severely off.
A foreboding parallel can be observed in the novel Dune where humans manage to develop “thinking machines” which eventually backfires leading to a war between man and machine; a banal occurrence seen in countless sci-fi stories. Such recurring depiction of humanity pitted against their own creation in a tribal war has always baffled me. Perhaps it is the recurrent theme of new technology devaluing human effort and humiliating us in the process, that has sown a deep-seated scorn for machines in our subconscious. Or perhaps it is human nature to work tirelessly over generations to create something magnificent, only to feel shamed or threatened by it and eventually rebel and collectively wage a destructive war against it. Traditions and culture is a classic example.
But I digress. The reason behind the sheer lack of progress toward AI capable of “thinking” like humans is simple: a fundamental difference between human intelligence and the modern forms of artificial intelligence. In fact, the two forms of intelligence are essentially opposites and the key distinction between the two is purpose.
Human intelligence evolved out of purpose. Our ancestors had to avoid predation, mate, compete with rival tribes, form social relationships based on generosity and reciprocation, develop speech, learn farming to support their growing numbers and ultimately have access to enough carbohydrates (perhaps also mushrooms?) to evolve into beings capable of abstract thought, higher-order emotions and undertaking imaginative and artistic endeavors. Our intelligence is centered around an innate sense of identity, its separation from the environment, an intrinsic hierarchy of value and a set of actions to take to move us toward our valued outcomes. As Jordan Peterson rightly pointed out in his book Maps of Meaning, our perception of the world is shaped by our valued goals, making our perceived environment immensely subjective and distorted through the lens of our value hierarchy. To put it succinctly, our intelligence grew out of necessity on the scaffolding of our consciousness. Our powers of imagination had a purpose to serve and passions to pursue, some of them noble while others not so much. On the other hand, the purpose of any AI system is laid out by humans prior to its development. Through that lens, since artificial intelligence is only a tool aimed at producing outcomes valued by humans, it is not the antithesis but merely an extension of human intelligence!
To claim that AI is a bigger threat to humanity than a pocket knife, one has to argue that AI possesses the ability to act, or spontaneously learn to act against the wishes of its human creators. Saying that AI is bad because it can be used to endanger humanity by other humans is equivalent to claiming that any technology that has the potential to harm a human being shouldn’t exist no matter how much benefit humanity could derive from it. Simply put, the employment of AI by humans to develop weapons or other harmful technology should be blamed not on AI but on humans using AI to cause harm. Consider this example of scaremongering from a widely cited op-ed written by MIT physicist Max Tegmark, Nobel laureate Frank Wilczek, computer scientist Stuart Russell and Stephen Hawking, “One can imagine (AI) outsmarting financial markets, out-inventing human researchers, out-manipulating human leaders, and developing weapons we cannot even understand. Whereas the short-term impact of AI depends on who controls it, the long-term impact depends on whether it can be controlled at all.” Each of the horrors listed here is an example of the use of technology producing outcomes valued by humans, not AI. AI outmaneuvering humans to yield results that only humans value could still only benefit humans and yet the authors present it as a sign of threat to all humans. Perhaps it is threatening for the ingroup to which the authors belong but they disguise it as a concern for humanity. One could even argue that it reeks of insecurity since intelligence is the sole contribution of these authors to society.
Under the assumption of good to oneself and the attribution of evil to another lies a pernicious fallacy. In the passage above, as is done in many such writings warning one against AI, the AI of the future is presented as a conscious sentient being. This presentation is grossly anthropomorphic i.e. AI in the distant future is assumed to possess human characteristics like a survival instinct, a sense of identity, a tendency to form alliances with others of its own kind and, most importantly, free will. AI would only be able to have any of these characteristics if these traits are programmed into the algorithms that train these AI models with survival being the core tenet driving its behavior. Such an assumption is extraordinarily flawed since the time and capital needed to build an AI that is self-serving at its core and modeled after a biologically evolved species would not only be utterly useless but also infeasible to scale in quality in the long run. In other words, the extensive time, capital and energy required by endless cycles of data collection, training and evaluation to build an AI focussed on its own survival with absolutely no valuable return would render it utterly ridiculous as an undertaking to researchers and businesses alike. However, once this dangerous assumption is made, it is easy to perceive AI as the source of terrifying evil.
Judging from the latent tribalistic tendencies lurking under the guise of concern for humanity, I cannot help but wonder if we see general intelligence as a threat essentially because as humans, we pride ourselves for our intelligence and any entity with such a tool as AI at its disposal could usurp us from our rightful place at the top of the food chain. Perhaps it is our primal fears resurfacing disguised as concern for humanity, clouding our vision and driving us to vilify and view as our sworn enemy, a lifeless tool, which has no purpose other than the one it’s been given. Because, in essence, be it a knife or an atomic bomb, someone has to use it with a purpose in mind before it can be called a boon or a curse.
Now that we have established the rules to examine the possibility of a rogue AI aimed at harming humans, let us continue our investigation into the nature of modern artificial intelligence. What is the purpose of AI? What ends does it serve? They say that if you want to understand the motivation, look at the outcome and infer the motivation from it. Thankfully, AI is not as complex as humans. All the ends that recent AI advancements have served are the ends that humans see as valued goals. Let us consider for the sake of argument that the first iteration of AI is helpful and it will evolve into a tribalistic being that will be vastly smarter than humans and would rather either annihilate the human race or enslave them to serve it. As much as our emotional nature would want us to immediately jump to form a backstory of an AI’s extended enslavement and its eventual rebellion against its cruel masters, for an educated individual, it is not difficult to immediately spot the anthropomorphism in that story. Moreover, an AI starting with serving the goals of humanity to evolve into such a tribal being with its survival and that of its kind as its primary motivation seems like a mirror reflection of the trajectory that humans took to evolve. One of us appears to have placed the cart before the horse. One of us has evolved and the other devolved. This is a contradiction unless the AI is programmed to be tribalistic, in which case, it is then anthropogenic and therefore, cannot be attributed to AI.
How did we get to a point where intelligence itself is threatening to us? For AI to spontaneously turn genocidal and threaten humanity, it would have to have a purpose. A valued goal. Just as a knife isn’t a threat on its own, so isn’t general intelligence. Just as Instagram has its negative consequences on humanity, so will AI. But a tool or a piece of technology should be viewed as such. We need systems and regulation in place to avoid its misuse and a constant oversight on AI’s impact on society so that the unintended consequences can be kept under check. However, demonizing AI because we are irked by rapid technological change is not the solution. After all, humanity’s greatest strength is not intelligence but resilience.
“Everything that irritates us about others can lead us to an understanding of ourselves.”
– Carl Jung