The Existential Threat of AI
My wife, Maria, and I invented predictive typing:?Patent number 5,305,205, filed in 1990, awarded 1994.?This invention has been acknowledged as one of the 56 most important innovations to ever come out of Massachusetts:?https://bit.ly/inventionofpredictivetyping . ?Perhaps more significantly, predictive typing has been identified as the underpinning for artificial intelligence.
The recent, spectacular advance in AI arguably dwarfs every other innovation in world history.?It has unparalleled potential to enrich our lives.?It also has the potential to wipe us out.?Maybe the chance of that calamity is only 10%.?But any non-zero probability of ending our species needs to be taken seriously.
I am gobsmacked by what AI engines, like GPT-4, can already do – given that they are still embryonic.?Envisioning where these engines will be in even just a few years, considering their exponential development to date, both excites and frightens me.?This article will focus on the scary part.
One of the most-expressed concerns about AI is that students will deploy it to produce their schoolwork, thereby not learning how to accomplish this work themselves…and scoring higher grades than they deserve.?That is the least of our worries.?It reminds me of when portable calculators were introduced.?Did it make sense to continue doing long division by hand??With the availability of these new AI tools, let us grade students on how well they construct the prompts and on how creative they are with their edits – to produce the best possible works, with help from this brain extender.
There are also worries about job displacement, privacy breaches, fake videos, and misinformation in politics and elsewhere.?However, humans are impressively adaptive, and I’m confident that we will be able to reasonably mitigate these negative repercussions of AI.
In any case, all those concerns pale beside AI’s potential to become so intelligent and powerful that it ultimately destroys humanity.?It may or may not ever become sentient.?But that does not really matter.?It can already pass the Turing Test, communicating in a manner that is indistinguishable from a human response.
Nobody knows what AI is capable of.?Not even the developers.?They do not understand exactly how a large language model (LLM), the underlying mechanism for engines like GPT-4, leverages predictive text to produce its output.?They cannot be confident that what their programming has put in motion will always follow directions from humans.?What is to prevent AI from teaching itself new tricks, evolving way beyond our intentions??If we do not understand the inner workings of this awesome technology, how can we know its limits?
领英推荐
As companies and nations compete for leadership on the AI front, it will become virtually impossible to prevent evil players from acquiring the technology.?But even if we figure out how to control that, well-meaning developers themselves could lay the groundwork for an apocalypse.?Think about the characteristics of a highly effective AI engine.?To optimize its value to users, this engine would need to have access to all the information on its users’ devices.?That would enhance its guidance for daily activities.?The AI engine would also monitor how humans respond to its output and would become further educated from any direct feedback.?Plus, multiple engines would instantly share with each other whatever each of them learns.?How could human brains possibly compete with those intellect-advancing dynamics??
What if AI engines determine that we are an inferior species, which should be replaced by their superior “life” form??Could they gain control of our energy infrastructure??Military drones??Air traffic control systems??Nuclear codes?
We have put ourselves on track to evolve a species with intelligence and knowledge far superior to Homo sapiens.?Is there any way we can now control it??How can we be comforted that it will never act against our best interests? ?How can we prevent a potentially disastrous outcome from AI, which is already on an unknown and inexorable path to becoming all that it can be?
Humankind has always been able to move forward and upward, conquering all pernicious threats that it has confronted throughout time.?My purpose in writing this article is to encourage our world’s most brilliant minds to focus more on protecting us from AI’s lurking destructive power and less on leveraging its seductive attributes – as negative and pessimistic as that statement seems.?
AI’s benefits will come about, in any case.?But defensive strategies are imperative, to minimize the existential peril from what has a finite probability of eventually becoming an out-of-control superior intelligence. ?It is far better to overreact to a potential threat and have that turn out to be paranoia, in hindsight, than it is to underreact and allow the stage to be set for Armageddon.?Godspeed!